May 29, 2025
Researchers describe a coding agent that changes its own software and tests candidate versions against coding benchmarks.
An evolving archive of agents explores alternative improvements, with benchmark evaluation rather than formal proof of general intelligence.